X-Git-Url: http://git.osdn.jp/view?a=blobdiff_plain;f=libhb%2Fsync.c;h=c391945b0483176230f6db565de9a1d7ad1a172c;hb=f9341345b37e0738a140423297c222a7e40eddab;hp=2a4818af30f8b6faf1e846a048a09be3c7e66eca;hpb=84a8d8155484a03a5db2cf41cc57e304ec578225;p=handbrake-jp%2Fhandbrake-jp-git.git diff --git a/libhb/sync.c b/libhb/sync.c index 2a4818af..c391945b 100644 --- a/libhb/sync.c +++ b/libhb/sync.c @@ -1,13 +1,13 @@ /* $Id: sync.c,v 1.38 2005/04/14 21:57:58 titer Exp $ This file is part of the HandBrake source code. - Homepage: . + Homepage: . It may be used under the terms of the GNU General Public License. */ #include "hb.h" - +#include "hbffmpeg.h" +#include #include "samplerate.h" -#include "ffmpeg/avcodec.h" #ifdef INT64_MIN #undef INT64_MIN /* Because it isn't defined correctly in Zeta */ @@ -22,10 +22,8 @@ typedef struct int64_t next_start; /* start time of next output frame */ int64_t next_pts; /* start time of next input frame */ - int64_t start_silence; /* if we're inserting silence, the time we started */ int64_t first_drop; /* PTS of first 'went backwards' frame dropped */ int drop_count; /* count of 'time went backwards' drops */ - int inserting_silence; /* Raw */ SRC_STATE * state; @@ -40,8 +38,11 @@ typedef struct struct hb_work_private_s { hb_job_t * job; - int done; - + int busy; // bitmask with one bit for each active input + // (bit 0 = video; 1 = audio 0, 2 = audio 1, ... + // appropriate bit is cleared when input gets + // an eof buf. syncWork returns done when all + // bits are clear. /* Video */ hb_subtitle_t * subtitle; int64_t pts_offset; @@ -49,9 +50,12 @@ struct hb_work_private_s int64_t next_pts; /* start time of next input frame */ int64_t first_drop; /* PTS of first 'went backwards' frame dropped */ int drop_count; /* count of 'time went backwards' drops */ + int drops; /* frames dropped to make a cbr video stream */ + int dups; /* frames duplicated to make a cbr video stream */ int video_sequence; int count_frames; int count_frames_max; + int chap_mark; /* to propagate chapter mark across a drop */ hb_buffer_t * cur; /* The next picture to process */ /* Audio */ @@ -67,9 +71,8 @@ struct hb_work_private_s * Local prototypes **********************************************************************/ static void InitAudio( hb_work_object_t * w, int i ); -static int SyncVideo( hb_work_object_t * w ); +static void SyncVideo( hb_work_object_t * w ); static void SyncAudio( hb_work_object_t * w, int i ); -static int NeedSilence( hb_work_object_t * w, hb_audio_t *, int i ); static void InsertSilence( hb_work_object_t * w, int i, int64_t d ); static void UpdateState( hb_work_object_t * w ); @@ -91,32 +94,41 @@ int syncInit( hb_work_object_t * w, hb_job_t * job ) pv->job = job; pv->pts_offset = INT64_MIN; - pv->count_frames = 0; /* Calculate how many video frames we are expecting */ - duration = 0; - for( i = job->chapter_start; i <= job->chapter_end; i++ ) + if (job->pts_to_stop) { - chapter = hb_list_item( title->list_chapter, i - 1 ); - duration += chapter->duration; + duration = job->pts_to_stop + 90000; } - duration += 90000; + else + { + duration = 0; + for( i = job->chapter_start; i <= job->chapter_end; i++ ) + { + chapter = hb_list_item( title->list_chapter, i - 1 ); + duration += chapter->duration; + } + duration += 90000; /* 1 second safety so we're sure we won't miss anything */ + } pv->count_frames_max = duration * job->vrate / job->vrate_base / 90000; hb_log( "sync: expecting %d video frames", pv->count_frames_max ); + pv->busy |= 1; /* Initialize libsamplerate for every audio track we have */ - for( i = 0; i < hb_list_count( title->list_audio ); i++ ) + if ( ! job->indepth_scan ) { - InitAudio( w, i ); + for( i = 0; i < hb_list_count( title->list_audio ); i++ ) + { + pv->busy |= ( 1 << (i + 1) ); + InitAudio( w, i ); + } } /* Get subtitle info, if any */ pv->subtitle = hb_list_item( title->list_subtitle, 0 ); - pv->video_sequence = 0; - return 0; } @@ -130,22 +142,26 @@ void syncClose( hb_work_object_t * w ) hb_work_private_t * pv = w->private_data; hb_job_t * job = pv->job; hb_title_t * title = job->title; - + hb_audio_t * audio = NULL; int i; - if( pv->cur ) hb_buffer_close( &pv->cur ); + if( pv->cur ) + { + hb_buffer_close( &pv->cur ); + } - for( i = 0; i < hb_list_count( title->list_audio ); i++ ) + hb_log( "sync: got %d frames, %d expected", + pv->count_frames, pv->count_frames_max ); + + if (pv->drops || pv->dups ) { - if ( pv->sync_audio[i].start_silence ) - { - hb_log( "sync: added %d ms of silence to audio %d", - (int)((pv->sync_audio[i].next_pts - - pv->sync_audio[i].start_silence) / 90), i ); - } + hb_log( "sync: %d frames dropped, %d duplicated", pv->drops, pv->dups ); + } - if( job->acodec & HB_ACODEC_AC3 || - job->audio_mixdowns[i] == HB_AMIXDOWN_AC3 ) + for( i = 0; i < hb_list_count( title->list_audio ); i++ ) + { + audio = hb_list_item( title->list_audio, i ); + if( audio->config.out.codec == HB_ACODEC_AC3 ) { free( pv->sync_audio[i].ac3_buf ); } @@ -175,17 +191,16 @@ int syncWork( hb_work_object_t * w, hb_buffer_t ** unused1, hb_work_private_t * pv = w->private_data; int i; - /* If we ever got a video frame, handle audio now */ - if( pv->pts_offset != INT64_MIN ) + if ( pv->busy & 1 ) + SyncVideo( w ); + + for( i = 0; i < hb_list_count( pv->job->title->list_audio ); i++ ) { - for( i = 0; i < hb_list_count( pv->job->title->list_audio ); i++ ) - { + if ( pv->busy & ( 1 << (i + 1) ) ) SyncAudio( w, i ); - } } - /* Handle video */ - return SyncVideo( w ); + return ( pv->busy? HB_WORK_OK : HB_WORK_DONE ); } hb_work_object_t hb_sync = @@ -207,8 +222,7 @@ static void InitAudio( hb_work_object_t * w, int i ) sync = &pv->sync_audio[i]; sync->audio = hb_list_item( title->list_audio, i ); - if( job->acodec & HB_ACODEC_AC3 || - job->audio_mixdowns[i] == HB_AMIXDOWN_AC3 ) + if( sync->audio->config.out.codec == HB_ACODEC_AC3 ) { /* Have a silent AC-3 frame ready in case we have to fill a gap */ @@ -219,11 +233,11 @@ static void InitAudio( hb_work_object_t * w, int i ) codec = avcodec_find_encoder( CODEC_ID_AC3 ); c = avcodec_alloc_context(); - c->bit_rate = sync->audio->bitrate; - c->sample_rate = sync->audio->rate; - c->channels = 2; + c->bit_rate = sync->audio->config.in.bitrate; + c->sample_rate = sync->audio->config.in.samplerate; + c->channels = HB_INPUT_CH_LAYOUT_GET_DISCRETE_COUNT( sync->audio->config.in.channel_layout ); - if( avcodec_open( c, codec ) < 0 ) + if( hb_avcodec_open( c, codec ) < 0 ) { hb_log( "sync: avcodec_open failed" ); return; @@ -231,8 +245,8 @@ static void InitAudio( hb_work_object_t * w, int i ) zeros = calloc( AC3_SAMPLES_PER_FRAME * sizeof( short ) * c->channels, 1 ); - sync->ac3_size = sync->audio->bitrate * AC3_SAMPLES_PER_FRAME / - sync->audio->rate / 8; + sync->ac3_size = sync->audio->config.in.bitrate * AC3_SAMPLES_PER_FRAME / + sync->audio->config.in.samplerate / 8; sync->ac3_buf = malloc( sync->ac3_size ); if( avcodec_encode_audio( c, sync->ac3_buf, sync->ac3_size, @@ -242,14 +256,14 @@ static void InitAudio( hb_work_object_t * w, int i ) } free( zeros ); - avcodec_close( c ); + hb_avcodec_close( c ); av_free( c ); } else { /* Initialize libsamplerate */ int error; - sync->state = src_new( SRC_LINEAR, HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(sync->audio->amixdown), &error ); + sync->state = src_new( SRC_SINC_MEDIUM_QUALITY, HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(sync->audio->config.out.mixdown), &error ); sync->data.end_of_input = 0; } } @@ -259,43 +273,25 @@ static void InitAudio( hb_work_object_t * w, int i ) *********************************************************************** * **********************************************************************/ -static int SyncVideo( hb_work_object_t * w ) +static void SyncVideo( hb_work_object_t * w ) { hb_work_private_t * pv = w->private_data; hb_buffer_t * cur, * next, * sub = NULL; hb_job_t * job = pv->job; - if( pv->done ) - { - return HB_WORK_DONE; - } - - if( hb_thread_has_exited( job->reader ) && - !hb_fifo_size( job->fifo_mpeg2 ) && - !hb_fifo_size( job->fifo_raw ) ) - { - /* All video data has been processed already, we won't get - more */ - hb_log( "sync: got %d frames, %d expected", - pv->count_frames, pv->count_frames_max ); - pv->done = 1; - - hb_buffer_t * buf_tmp; - - // Drop an empty buffer into our output to ensure that things - // get flushed all the way out. - buf_tmp = hb_buffer_init(0); // Empty end buffer - hb_fifo_push( job->fifo_sync, buf_tmp ); - - return HB_WORK_DONE; - } - if( !pv->cur && !( pv->cur = hb_fifo_get( job->fifo_raw ) ) ) { /* We haven't even got a frame yet */ - return HB_WORK_OK; + return; } cur = pv->cur; + if( cur->size == 0 ) + { + /* we got an end-of-stream. Feed it downstream & signal that we're done. */ + hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) ); + pv->busy &=~ 1; + return; + } /* At this point we have a frame to process. Let's check 1) if we will be able to push into the fifo ahead @@ -306,6 +302,16 @@ static int SyncVideo( hb_work_object_t * w ) { hb_buffer_t * buf_tmp; + if( next->size == 0 ) + { + /* we got an end-of-stream. Feed it downstream & signal that + * we're done. Note that this means we drop the final frame of + * video (we don't know its duration). On DVDs the final frame + * is often strange and dropping it seems to be a good idea. */ + hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) ); + pv->busy &=~ 1; + return; + } if( pv->pts_offset == INT64_MIN ) { /* This is our first frame */ @@ -325,6 +331,10 @@ static int SyncVideo( hb_work_object_t * w ) } } + if( cur->new_chap ) { + hb_log("sync got new chapter %d", cur->new_chap ); + } + /* * since the first frame is always 0 and the upstream reader code * is taking care of adjusting for pts discontinuities, we just have @@ -336,7 +346,7 @@ static int SyncVideo( hb_work_object_t * w ) * can deal with overlaps of up to a frame time but anything larger * we handle by dropping frames here. */ - if ( pv->next_pts - next->start > 1000 ) + if ( (int64_t)( next->start - cur->start ) <= 0 ) { if ( pv->first_drop == 0 ) { @@ -344,15 +354,20 @@ static int SyncVideo( hb_work_object_t * w ) } ++pv->drop_count; buf_tmp = hb_fifo_get( job->fifo_raw ); + if ( buf_tmp->new_chap ) + { + // don't drop a chapter mark when we drop the buffer + pv->chap_mark = buf_tmp->new_chap; + } hb_buffer_close( &buf_tmp ); continue; } if ( pv->first_drop ) { - hb_log( "sync: video time went backwards %d ms, dropped %d frames " - "(frame %lld, expected %lld)", - (int)( pv->next_pts - pv->first_drop ) / 90, pv->drop_count, - pv->first_drop, pv->next_pts ); + hb_log( "sync: video time didn't advance - dropped %d frames " + "(delta %d ms, current %lld, next %lld, dur %d)", + pv->drop_count, (int)( cur->start - pv->first_drop ) / 90, + cur->start, next->start, (int)( next->start - cur->start ) ); pv->first_drop = 0; pv->drop_count = 0; } @@ -432,9 +447,9 @@ static int SyncVideo( hb_work_object_t * w ) * Subtitle is on for less than three seconds, extend * the time that it is displayed to make it easier * to read. Make it 3 seconds or until the next - * subtitle is displayed. + * subtitle is displayed. * - * This is in response to Indochine which only + * This is in response to Indochine which only * displays subs for 1 second - too fast to read. */ sub->stop = sub->start + ( 3 * 90000 ); @@ -496,27 +511,99 @@ static int SyncVideo( hb_work_object_t * w ) } } - /* - * Adjust the pts of the current frame so that it's contiguous - * with the previous frame. The start time of the current frame - * has to be the end time of the previous frame and the stop - * time has to be the start of the next frame. We don't - * make any adjustments to the source timestamps other than removing - * the clock offsets (which also removes pts discontinuities). - * This means we automatically encode at the source's frame rate. - * MP2 uses an implicit duration (frames end when the next frame - * starts) but more advanced containers like MP4 use an explicit - * duration. Since we're looking ahead one frame we set the - * explicit stop time from the start time of the next frame. - */ - buf_tmp = cur; - pv->cur = cur = hb_fifo_get( job->fifo_raw ); - pv->next_pts = next->start; - int64_t duration = next->start - buf_tmp->start; + int64_t duration; + if ( job->mux & HB_MUX_AVI || job->cfr ) + { + /* + * The concept of variable frame rate video was a bit too advanced + * for Microsoft so AVI doesn't support it. Since almost all dvd + * video is VFR we have to convert it to constant frame rate to + * put it in an AVI container. So here we duplicate, drop and + * otherwise trash video frames to appease the gods of Redmond. + */ + + /* mpeg durations are exact when expressed in ticks of the + * 27MHz System clock but not in HB's 90KHz PTS clock. To avoid + * a truncation bias that will eventually cause the audio to desync + * we compute the duration of the next frame using 27MHz ticks + * then truncate it to 90KHz. */ + duration = ( (int64_t)(pv->count_frames + 1 ) * job->vrate_base ) / 300 - + pv->next_start; + + /* We don't want the input & output clocks to be exactly in phase + * otherwise small variations in the time will cause us to think + * we're a full frame off & there will be lots of drops and dups. + * We offset the input clock by half the duration so it's maximally + * out of phase with the output clock. */ + if( cur->start < pv->next_start - ( duration >> 1 ) ) + { + /* current frame too old - drop it */ + if ( cur->new_chap ) + { + pv->chap_mark = cur->new_chap; + } + hb_buffer_close( &cur ); + pv->cur = cur = hb_fifo_get( job->fifo_raw ); + pv->next_pts = next->start; + ++pv->drops; + continue; + } + + if( next->start > pv->next_start + duration + ( duration >> 1 ) ) + { + /* next frame too far ahead - dup current frame */ + buf_tmp = hb_buffer_init( cur->size ); + hb_buffer_copy_settings( buf_tmp, cur ); + memcpy( buf_tmp->data, cur->data, cur->size ); + buf_tmp->sequence = cur->sequence; + ++pv->dups; + } + else + { + /* this frame in our time window & doesn't need to be duped */ + buf_tmp = cur; + pv->cur = cur = hb_fifo_get( job->fifo_raw ); + pv->next_pts = next->start; + } + } + else + { + /* + * Adjust the pts of the current frame so that it's contiguous + * with the previous frame. The start time of the current frame + * has to be the end time of the previous frame and the stop + * time has to be the start of the next frame. We don't + * make any adjustments to the source timestamps other than removing + * the clock offsets (which also removes pts discontinuities). + * This means we automatically encode at the source's frame rate. + * MP2 uses an implicit duration (frames end when the next frame + * starts) but more advanced containers like MP4 use an explicit + * duration. Since we're looking ahead one frame we set the + * explicit stop time from the start time of the next frame. + */ + buf_tmp = cur; + pv->cur = cur = hb_fifo_get( job->fifo_raw ); + pv->next_pts = cur->start; + duration = cur->start - buf_tmp->start; + if ( duration <= 0 ) + { + hb_log( "sync: invalid video duration %lld, start %lld, next %lld", + duration, buf_tmp->start, next->start ); + } + } + buf_tmp->start = pv->next_start; pv->next_start += duration; buf_tmp->stop = pv->next_start; + if ( pv->chap_mark ) + { + // we have a pending chapter mark from a recent drop - put it on this + // buffer (this may make it one frame late but we can't do any better). + buf_tmp->new_chap = pv->chap_mark; + pv->chap_mark = 0; + } + /* If we have a subtitle for this picture, copy it */ /* FIXME: we should avoid this memcpy */ if( sub ) @@ -538,19 +625,16 @@ static int SyncVideo( hb_work_object_t * w ) /* Make sure we won't get more frames then expected */ if( pv->count_frames >= pv->count_frames_max * 2) { - hb_log( "sync: got too many frames (%d), exiting early", pv->count_frames ); - pv->done = 1; + hb_log( "sync: got too many frames (%d), exiting early", + pv->count_frames ); - // Drop an empty buffer into our output to ensure that things - // get flushed all the way out. - buf_tmp = hb_buffer_init(0); // Empty end buffer - hb_fifo_push( job->fifo_sync, buf_tmp ); - - break; + // Drop an empty buffer into our output to ensure that things + // get flushed all the way out. + hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) ); + pv->busy &=~ 1; + return; } } - - return HB_WORK_OK; } static void OutputAudioFrame( hb_job_t *job, hb_audio_t *audio, hb_buffer_t *buf, @@ -558,25 +642,15 @@ static void OutputAudioFrame( hb_job_t *job, hb_audio_t *audio, hb_buffer_t *buf { int64_t start = sync->next_start; int64_t duration = buf->stop - buf->start; - if (duration <= 0 || - duration > ( 90000 * AC3_SAMPLES_PER_FRAME ) / audio->rate ) - { - hb_log("sync: audio %d weird duration %lld, start %lld, stop %lld, next %lld", - i, duration, buf->start, buf->stop, sync->next_pts); - if ( duration <= 0 ) - { - duration = ( 90000 * AC3_SAMPLES_PER_FRAME ) / audio->rate; - buf->stop = buf->start + duration; - } - } + sync->next_pts += duration; - if( /* audio->rate == job->arate || This should work but doesn't */ - job->acodec & HB_ACODEC_AC3 || - job->audio_mixdowns[i] == HB_AMIXDOWN_AC3 ) + if( audio->config.in.samplerate == audio->config.out.samplerate || + audio->config.out.codec == HB_ACODEC_AC3 || + audio->config.out.codec == HB_ACODEC_DCA ) { /* - * If we don't have to do sample rate conversion or this audio is AC3 + * If we don't have to do sample rate conversion or this audio is * pass-thru just send the input buffer downstream after adjusting * its timestamps to make the output stream continuous. */ @@ -586,15 +660,25 @@ static void OutputAudioFrame( hb_job_t *job, hb_audio_t *audio, hb_buffer_t *buf /* Not pass-thru - do sample rate conversion */ int count_in, count_out; hb_buffer_t * buf_raw = buf; - int channel_count = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->amixdown) * + int channel_count = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->config.out.mixdown) * sizeof( float ); count_in = buf_raw->size / channel_count; - count_out = ( buf_raw->stop - buf_raw->start ) * job->arate / 90000; + /* + * When using stupid rates like 44.1 there will always be some + * truncation error. E.g., a 1536 sample AC3 frame will turn into a + * 1536*44.1/48.0 = 1411.2 sample frame. If we just truncate the .2 + * the error will build up over time and eventually the audio will + * substantially lag the video. libsamplerate will keep track of the + * fractional sample & give it to us when appropriate if we give it + * an extra sample of space in the output buffer. + */ + count_out = ( duration * audio->config.out.samplerate ) / 90000 + 1; sync->data.input_frames = count_in; sync->data.output_frames = count_out; - sync->data.src_ratio = (double)count_out / (double)count_in; + sync->data.src_ratio = (double)audio->config.out.samplerate / + (double)audio->config.in.samplerate; buf = hb_buffer_init( count_out * channel_count ); sync->data.data_in = (float *) buf_raw->data; @@ -607,10 +691,12 @@ static void OutputAudioFrame( hb_job_t *job, hb_audio_t *audio, hb_buffer_t *buf hb_buffer_close( &buf_raw ); buf->size = sync->data.output_frames_gen * channel_count; + duration = ( sync->data.output_frames_gen * 90000 ) / + audio->config.out.samplerate; } + buf->frametype = HB_FRAME_AUDIO; buf->start = start; buf->stop = start + duration; - buf->frametype = HB_FRAME_AUDIO; sync->next_start = start + duration; hb_fifo_push( fifo, buf ); } @@ -628,82 +714,82 @@ static void SyncAudio( hb_work_object_t * w, int i ) hb_audio_t * audio = sync->audio; hb_buffer_t * buf; hb_fifo_t * fifo; - int rate; - if( job->acodec & HB_ACODEC_AC3 || - job->audio_mixdowns[i] == HB_AMIXDOWN_AC3 ) + if( audio->config.out.codec == HB_ACODEC_AC3 ) { - fifo = audio->fifo_out; - rate = audio->rate; + fifo = audio->priv.fifo_out; } else { - fifo = audio->fifo_sync; - rate = job->arate; + fifo = audio->priv.fifo_sync; } - while( !hb_fifo_is_full( fifo ) && ( buf = hb_fifo_see( audio->fifo_raw ) ) ) + while( !hb_fifo_is_full( fifo ) && ( buf = hb_fifo_see( audio->priv.fifo_raw ) ) ) { - if ( sync->next_pts - buf->start > 500 ) + /* if the next buffer is an eof send it downstream */ + if ( buf->size <= 0 ) { - /* - * audio time went backwards by more than a frame time (this can - * happen when we reset the PTS because of lost data). - * Discard data that's in the past. - */ - if ( sync->first_drop == 0 ) + buf = hb_fifo_get( audio->priv.fifo_raw ); + hb_fifo_push( fifo, buf ); + pv->busy &=~ (1 << (i + 1) ); + return; + } + if ( (int64_t)( buf->start - sync->next_pts ) < 0 ) + { + // audio time went backwards. + // If our output clock is more than a half frame ahead of the + // input clock drop this frame to move closer to sync. + // Otherwise drop frames until the input clock matches the output clock. + if ( sync->first_drop || sync->next_start - buf->start > 90*15 ) { - sync->first_drop = buf->start; + // Discard data that's in the past. + if ( sync->first_drop == 0 ) + { + sync->first_drop = sync->next_pts; + } + ++sync->drop_count; + buf = hb_fifo_get( audio->priv.fifo_raw ); + hb_buffer_close( &buf ); + continue; } - ++sync->drop_count; - buf = hb_fifo_get( audio->fifo_raw ); - hb_buffer_close( &buf ); - continue; + sync->next_pts = buf->start; } if ( sync->first_drop ) { + // we were dropping old data but input buf time is now current hb_log( "sync: audio %d time went backwards %d ms, dropped %d frames " - "(frame %lld, expected %lld)", i, + "(next %lld, current %lld)", i, (int)( sync->next_pts - sync->first_drop ) / 90, sync->drop_count, sync->first_drop, sync->next_pts ); sync->first_drop = 0; sync->drop_count = 0; + sync->next_pts = buf->start; } - - if ( sync->inserting_silence && buf->start - sync->next_pts > 0 ) + if ( buf->start - sync->next_pts >= (90 * 70) ) { - /* - * if we're within one frame time of the amount of silence - * we need, insert just what we need otherwise insert a frame time. - */ - int64_t framedur = buf->stop - buf->start; - if ( buf->start - sync->next_pts <= framedur ) + if ( buf->start - sync->next_pts > (90000LL * 60) ) { - InsertSilence( w, i, buf->start - sync->next_pts ); - sync->inserting_silence = 0; - } - else - { - InsertSilence( w, i, framedur ); + // there's a gap of more than a minute between the last + // frame and this. assume we got a corrupted timestamp + // and just drop the next buf. + hb_log( "sync: %d minute time gap in audio %d - dropping buf" + " start %lld, next %lld", + (int)((buf->start - sync->next_pts) / (90000*60)), + i, buf->start, sync->next_pts ); + buf = hb_fifo_get( audio->priv.fifo_raw ); + hb_buffer_close( &buf ); + continue; } - continue; - } - if ( buf->start - sync->next_pts >= (90 * 100) ) - { /* - * there's a gap of at least 100ms between the last + * there's a gap of at least 70ms between the last * frame we processed & the next. Fill it with silence. */ - if ( ! sync->inserting_silence ) - { - hb_log( "sync: adding %d ms of silence to audio %d" - " start %lld, next %lld", - (int)((buf->start - sync->next_pts) / 90), - i, buf->start, sync->next_pts ); - sync->inserting_silence = 1; - } - InsertSilence( w, i, buf->stop - buf->start ); - continue; + hb_log( "sync: adding %d ms of silence to audio %d" + " start %lld, next %lld", + (int)((buf->start - sync->next_pts) / 90), + i, buf->start, sync->next_pts ); + InsertSilence( w, i, buf->start - sync->next_pts ); + return; } /* @@ -711,45 +797,9 @@ static void SyncAudio( hb_work_object_t * w, int i ) * audio stream and are ready to inject the next input frame into * the output stream. */ - buf = hb_fifo_get( audio->fifo_raw ); + buf = hb_fifo_get( audio->priv.fifo_raw ); OutputAudioFrame( job, audio, buf, sync, fifo, i ); } - - if( NeedSilence( w, audio, i ) ) - { - InsertSilence( w, i, (90000 * AC3_SAMPLES_PER_FRAME) / sync->audio->rate ); - } -} - -static int NeedSilence( hb_work_object_t * w, hb_audio_t * audio, int i ) -{ - hb_work_private_t * pv = w->private_data; - hb_job_t * job = pv->job; - hb_sync_audio_t * sync = &pv->sync_audio[i]; - - if( hb_fifo_size( audio->fifo_in ) || - hb_fifo_size( audio->fifo_raw ) || - hb_fifo_size( audio->fifo_sync ) || - hb_fifo_size( audio->fifo_out ) ) - { - /* We have some audio, we are fine */ - return 0; - } - - /* No audio left in fifos */ - - if( hb_thread_has_exited( job->reader ) ) - { - /* We might miss some audio to complete encoding and muxing - the video track */ - if ( sync->start_silence == 0 ) - { - hb_log("sync: reader has exited, adding silence to audio %d", i); - sync->start_silence = sync->next_pts; - } - return 1; - } - return 0; } static void InsertSilence( hb_work_object_t * w, int i, int64_t duration ) @@ -758,23 +808,37 @@ static void InsertSilence( hb_work_object_t * w, int i, int64_t duration ) hb_job_t *job = pv->job; hb_sync_audio_t *sync = &pv->sync_audio[i]; hb_buffer_t *buf; + hb_fifo_t *fifo; - if( job->acodec & HB_ACODEC_AC3 || job->audio_mixdowns[i] == HB_AMIXDOWN_AC3 ) - { - buf = hb_buffer_init( sync->ac3_size ); - buf->start = sync->next_pts; - buf->stop = buf->start + duration; - memcpy( buf->data, sync->ac3_buf, buf->size ); - OutputAudioFrame( job, sync->audio, buf, sync, sync->audio->fifo_out, i ); - } - else + // to keep pass-thru and regular audio in sync we generate silence in + // AC3 frame-sized units. If the silence duration isn't an integer multiple + // of the AC3 frame duration we will truncate or round up depending on + // which minimizes the timing error. + const int frame_dur = ( 90000 * AC3_SAMPLES_PER_FRAME ) / + sync->audio->config.in.samplerate; + int frame_count = ( duration + (frame_dur >> 1) ) / frame_dur; + + while ( --frame_count >= 0 ) { - buf = hb_buffer_init( duration * sizeof( float ) * - HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(sync->audio->amixdown) ); - buf->start = sync->next_pts; - buf->stop = buf->start + duration; - memset( buf->data, 0, buf->size ); - OutputAudioFrame( job, sync->audio, buf, sync, sync->audio->fifo_sync, i ); + if( sync->audio->config.out.codec == HB_ACODEC_AC3 ) + { + buf = hb_buffer_init( sync->ac3_size ); + buf->start = sync->next_pts; + buf->stop = buf->start + frame_dur; + memcpy( buf->data, sync->ac3_buf, buf->size ); + fifo = sync->audio->priv.fifo_out; + } + else + { + buf = hb_buffer_init( AC3_SAMPLES_PER_FRAME * sizeof( float ) * + HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT( + sync->audio->config.out.mixdown) ); + buf->start = sync->next_pts; + buf->stop = buf->start + frame_dur; + memset( buf->data, 0, buf->size ); + fifo = sync->audio->priv.fifo_sync; + } + OutputAudioFrame( job, sync->audio, buf, sync, fifo, i ); } }